3D installation prevents reboot into desktop

[lothar2]

Armbian 22.11.2 Cinnamon desktop was successfully installed onto eMMc storage from a microSD card. However, after installing the 3D accelleration following the official Armbian instructions, the Rock 5b failed to load the desktop. Instead, after the display of the Armbian splash screen, the screen turns black with a flashing white cursor. This problem started on 25 February. Before that day, 22.11.2 Cinnamon desktop successfully operated with 3D accelleration for two months.

 

The following was used to install the 3D accelleration:

sudo add-apt-repository ppa:liujianfeng1994/panfork-mesa

sudo add-apt-repository ppa:liujianfeng1994/rockchip-multimedia

sudo apt update

sudo apt dist-upgrade

sudo apt install kodi

 

The problem appears exactly after the entry of sudo apt dist-upgrade and reboot of the OS. Each step before dist-upgrade was tested by executing the command and then rebooting the OS. The desktop always appeared.

Armbian 23.02 Cinnamon desktop was also installed but the same problem emerged after the installation of 3D accelleration. It is possible to launch the CLI at the blinking cursor stage by pressing down control + alt + F2 at the same time. BASH commands can then be executed as normal. I did not try to launch the desktop because I didn't know how best to proceed. Can anyone advise on how to move forward with this problem? Thank you for your help :^)

[specs]

if you still have access to a prompt I'd suggest you login and remove the iujianfeng1994-ubuntu-panfork-mesa-kinetic.list from /etc/apt/sources.list.d/

 

What I did:

• run "dpkg --get-selections > installed-pkgs"
• remove everything which includes panfork in the name with dpkg, manually
• remove the liujianfeng ppa
• run "apt clean"
• run "apt update"
• reinstall the desktop
• reinstall all the pkgs removed when removing the panfork packages (I used ""dpkg --get-selections > installed2" and diff)
• reboot and the desktop reappears

 

The sunny side is, removing panfork means some weird bug where the pointer disappears after the screensaver kicks in disappeared.

The dark side is, you lose hardware GPU acceleration.

 

Alternative solution (untested):

• update mesa23.0~panfork~csf to git221210.120202c6757~j+1   JianFeng Liu (10 hours ago) <= wasn't there this weekend.

 

Note that some developpers ban the panfork project, since I had the same issue (and the above bug) I dumped it for now.

I'll wait to see what mesa driver gets adopted in the official armbian distribution.

[DIV_ZERO]

I compiled "mesa23.0~panfork~csf to git221210.120202c6757~j+1" myself because the JianFeng Liu package is not available in the unsupported Sid + Xfce flavour of Armbian. It's working fine since weeks (instructions for compile: https://gitlab.com/panfork/mesa). At the moment there is still no support in upstream mesa for the G610. I see some commits, but looks like it's still a very long way to go. Btw, I found this topic in the mesa issue tracker: https://gitlab.freedesktop.org/mesa/mesa/-/issues/8054 which sounds not promising.

[lothar2]

Thank you to Specs for proposing some steps to resurrect my Cinnamon desktop. I did succeed in removing the panfork-mesa-kinetic.list from my sources list. However I did not find any files that contained the words panfork or mesa using dpkg and find. So I executed all the other steps and rebooted only to find that the desktop still did not appear. But the story did not end there…

 

Whilst removing the ppa source, the following message appeared: “You need to install mali-g610-firmware if you don’t have file /lib/firmware/mali.csffw.bin”

 

I checked that directory and indeed there was no file by that name. So, using the clues offered by Specs and DIV_ZERO, I found the file within a driver package at the following link:

 

https://launchpad.net/~liujianfeng1994/+archive/ubuntu/panfork-mesa/+packages

 

I copied and pasted the file into that location, rebooted, and the Cinnamon desktop reappeared. Using htop I noticed that the CPU load values were much lower than the non-accelerated desktop when I moved open windows on the desktop and streamed videos from YouTube. < glxinfo -B > indicated that the GPU was accelerated. How did that happen?!? :^)

 

Although the quality of YouTube streaming is not as good as before, I can live with these limitations until a more stable version of Mali g610 firmware is available. I am new to this aspect of the SBC game, and this experience has made me wonder (and I am not complaining out loud)…Why is this GPU driver still in development? The RK3588 was released over a year ago and judging by its quick adoption by board makers appears to be another international bestseller. I have been using the RK3399 for over four years and I have not had any device support problems. I have poured over forum discussions and manufacturer docs and still don’t have any sense about how this situation came about.

 

Can anyone fill me in on the Mali g610 backstory? :^)

[linhz0hz]

I  thought the RK3399 GPU drivers were included in the mainline very very recently? You might not notice it because it was software rendered?

[DIV_ZERO]

Today I found this in the "panfrost" irc channel:

Zitat

10:16 <Googulator> This is the latest pancsf code, right? https://gitlab.freedesktop.org/bbrezillon/linux/-/tree/pancsf
10:34 Danct12 is now known as Guest7036
13:28 <Googulator> Testing the latest pancsf by @bbrezillon on top of a 6.2 kernel with drivers for the Rock 5B, plus the recently posted Mesa code. When I run glmark2-es2 (x11 version), I see heavy glitching during the jellyfish, terrain and refract tests, with this error printed to dmesg: https://pastebin.com/W2CSRGc6 - is this a known issue?
13:28 <Googulator> (The desktop itself is stable, after merging in recent pancsf kernel driver changes.)
13:48 <Googulator> The WARN stack traces are not part of the normal output of pancsf, I added them to the code for debugging.

Looks like there is some progress. However, sounds still rather experimental to me.